cadmium

Cadmium is a Natural Language Processing (NLP) library for Crystal. Included are classes and modules for tokenizing, inflecting, stemming, and creating n-grams with much more to come.

It's still in early development, but tests are being written as I go so hopefully it will be somewhat stable.

This library is heavily based on the natural library for node.js, and as such you can expect the API's to be very similar. As a point of fact, most of the specs for Cadmium were copied directly from natural and lightly modified.

Any utilities that can be internationalized will be eventually. For now English is the primary concern.

For full API documentation check out the docs.

Table of Contents

• Table of Contents
• Installation
• Usage
  • Tokenizers
    • Aggressive Tokenizer
    • Case Tokenizer
    • Regex Tokenizer
    • Treebank Word Tokenizer
    • Pragmatic Tokenizer
  • String Distance
    • Jaro-Winkler
    • Levenshtein
    • Pair
  • Stemmers
  • Phonetics
  • Inflectors
    • Nouns
    • Present Tense Verbs
    • Numbers
  • N-Grams
    • bigrams
    • trigrams
    • arbitrary n-grams
    • padding
  • Classifiers
  • tf-idf
  • Transliterator
  • Sentiment Analysis
  • Tries
  • Edge Weighted Digraph
  • Readability
  • WordNet
• Roadmap
• Contributing
• Contributors

Installation

Add this to your application's shard.yml:

dependencies:
  cadmium:
    github: watzon/cadmium
    branch: master

Usage

Require the cadmium library in your project

require "cadmium"

Tokenizers

Cadmium includes several different tokenizers, each of which is useful for different applications.

Aggressive Tokenizer

The aggressive tokenizer currently has localization available for:

• English (:en)
• Spanish (:es)
• Persian (:fa)
• French (:fr)
• Indonesian (:id)
• Dutch (:nl)
• Norwegian (:no)
• Polish (:pl)
• Portuguese (:pt)
• Russian (:ru)
• Serbian (:sb)
• Ukranian (:uk)
• Bulgarian (:bg)
• Swedish (:sv)

If no language is included it will default to English.

Use it like so:

tokenizer = Cadmium.aggressive_tokenizer.new(lang: :es)
tokenizer.tokenize("hola yo me llamo eduardo y esudié ingeniería")
# => ["hola", "yo", "me", "llamo", "eduardo", "y", "esudié", "ingeniería"]

Case Tokenizer

The case tokenizer doesn't rely on Regex and as such should be pretty fast. It should also work on an international basis fairly easily.

tokenizer = Cadmium.case_tokenizer.new
tokenizer.tokenize("these are strings")
# => ["these", "are", "strings"]

tokenizer = Cadmium.case_tokenizer.new(preserve_apostrophes: true)
tokenizer.tokenize("Affectueusement surnommé « Gabo » dans toute l'Amérique latine")
# => ["Affectueusement", "surnommé", "Gabo", "dans", "toute", "l", "Amérique", "latine"]

Regex Tokenizer

The whitespace tokenizer, word punctuation tokenizer, and word tokenizer all extend the regex tokenizer. It uses Regex to match on the correct values.

tokenizer = Cadmium.word_punctuation_tokenizer.new
tokenizer.tokenize("my dog hasn't any fleas.")
# => ["my", "dog", "hasn", "'", "t", "any", "fleas", "."]

Treebank Word Tokenizer

The treebank tokenizer uses regular expressions to tokenize text as in Penn Treebank. This implementation is a port of the tokenizer sed script written by Robert McIntyre. To read about treebanks you can visit wikipedia.

tokenizer = Cadmium.treebank_word_tokenizer.new
tokenizer.tokenize("If we 'all' can't go. I'll stay home.")
# => ["If", "we", "'all", "'", "ca", "n't", "go.", "I", "'ll", "stay", "home", "."]

Pragmatic Tokenizer

The pragmatic tokenizer is based off of the ruby gem from diasks2 which you can find here. It is a multilengual tokenizer which provides a wide array of options for tokenizing strings. For complete documentation check here.

Example is taken directly from the diasks2/pragmatic_tokenizer documentation, with a few modifications. Currently supported languages are:

• English (:en)
• Deutsch (:de)
• Czech (:cz)
• Bulgarian (:bg)
• Spanish (:sp)
• Portuguese (:pt)

text = "\"I said, 'what're you? Crazy?'\" said Sandowsky. \"I can't afford to do that.\""

Cadmium.pragmatic_tokenizer.new.tokenize(text)
# => ["\"", "i", "said", ",", "'", "what're", "you", "?", "crazy", "?", "'", "\"", "said", "sandowsky", ".", "\"", "i", "can't", "afford", "to", "do", "that", ".", "\""]

The initializer accepts the following options:

language:            :en, # the language of the string you are tokenizing
abbreviations:       Set{"a.b", "a"}, # a user-supplied array of abbreviations (downcased with ending period removed)
stop_words:          Set{"is", "the"}, # a user-supplied array of stop words (downcased)
remove_stop_words:   true, # remove stop words
contractions:        { "i'm" => "i am" }, # a user-supplied hash of contractions (key is the contracted form; value is the expanded                                             form - both the key and value should be downcased)
expand_contractions: true, # (i.e. ["isn't"] will change to two tokens ["is", "not"])
filter_languages:    [:en, :de], # process abbreviations, contractions and stop words for this array of languages
punctuation:         :none, # see below for more details
numbers:             :none, # see below for more details
remove_emoji:        true, # remove any emoji tokens
remove_urls:         true, # remove any urls
remove_emails:       true, # remove any emails
remove_domains:      true, # remove any domains
hashtags:            :keep_and_clean, # remove the hastag prefix
mentions:            :keep_and_clean, # remove the @ prefix
clean:               true, # remove some special characters
classic_filter:      true, # removes dots from acronyms and 's from the end of tokens
downcase:            false, # do not downcase tokens
minimum_length:      3, # remove any tokens less than 3 characters
long_word_split:     10 # split tokens longer than 10 characters at hypens or underscores

String Distance

Corundum provides an implimentation of two different string distance algorithms, the Jaro-Winkler Distance Algorithm and the Levenshtein Distance Algorithm.

Jaro-Winkler

The Jaro-Winkler algorithm returns a number between 0 and 1 which tells how closely two strings match (1 being perfect and 0 being not at all).

jwd = Cadmium.jaro_winkler.new

jwd.distance("dixon","dicksonx")
# => 0.8133333333333332

jwd.distance("same","same")
# => 1

jwd.distance("not","same")
# => 0.0

Levenshtein

The Levenshtein distance algorithm returns the number of edits (insertions, modifications, or deletions) required to transform one string into another.

Cadmium.levenshtein.distance("doctor", "doktor")
# => 1

Cadmium.levenshtein.distance("doctor", "doctor")
# => 0

Cadmium.levenshtein.distance("flad", "flaten")
# => 3

Pair

Pair Distance uses arbitrary n-grams to calculate how similar one string is to another. By calculating the bi-grams for a string, the pair distance algorithm first checks how many occurrences of each bi-gram occur in both strings, then it calculates their similarity with the formula simularity = (2 · intersections) / (s1size + s2size).

Cadmium.pair_distance.distance("night", "nacht")
# => 0.25

Stemmers

Currently Cadmium only comes with a Porter Stemmer, but Lancaster will be added soon. Stemmer methods stem and tokenize_and_stem have also been added to the String class to simplify use.

"words".stem
# => word

"i am waking up to the sounds of chainsaws".tokenize_and_stem
# => ["wake", "sound", "chainsaw"]

Phonetics

Phonetic matching (sounds-like) matching can be done with the SoundEx or Metaphone algorithms

sound_ex = Cadmium.sound_ex
metaphone = Cadmium.metaphone

sound_ex.process("phonetics")
# => "P532"

sound_ex.tokenize_and_phoneticize("Ruby aint got nothing on Crystal")
# => ["R100", "A530", "G300", "C234"]

# Keep word stops
sound_ex.tokenize_and_phoneticize("Ruby aint got nothing on Crystal", true)
# => ["R100", "A530", "G300", "N352", "O000", "C234"]

sound_ex.compare("phonetix", "phonetics")
# => true

metaphone.process("phonetics")
# => "FNTKS"

metaphone.tokenize_and_phoneticize("Ruby aint got nothing on Crystal")
# => ["RB", "ANT", "KT", "KRSTL"]

# Keep word stops
metaphone.tokenize_and_phoneticize("Ruby aint got nothing on Crystal", true)
# => ["RB", "ANT", "KT", "N0NK", "ON", "KRSTL"]

metaphone.compare("phonetix", "phonetics")
# => true

Both classes can also be used with attached String methods. The default class for String methods is Metaphone. The attached methods are phonetics, sounds_like, and tokenize_and_phoneticize.

"Crystal".phonetics
# => "KRSTL"

"Crystal".sounds_like("Krystal")
# => true

"Crystal".phonetics(nil, Cadmium::SoundEx)
# => "C234"

# Using a max length
"Constitution".phonetics(6, Cadmium::SoundEx)
# => "C52333"

Inflectors

Nouns

Nouns can be inflected using the NounInflector which has also been attached to the String class.

inflector = Cadmium.noun_inflector.new

inflector.pluralize("radius")
# => radii

inflector.singularize("radii")
# => radius

"person".pluralize
# => people

"people".singularize
# => person

Present Tense Verbs

Present tense verbs can be inflected with the PresentVerbInflector. This has also been attached to the string class.

inflector = Cadmium.present_verb_inflector.new

inflector.singularize("become")
# => became

inflector.pluralize("became")
# => become

"walk".singularize(false) # noun: false
# => walks

"walks".pluralize(false)  # noun: false
# => walk

Numbers

Numbers can be inflected with the CountInflector which also adds a method to_nth to the Int class.

Cadmium.count_inflector.nth(1)
# => 1st

Cadmium.count_inflector.nth(111)
# => 111th

153.to_nth
# => 153rd

N-Grams

N-Grams can be obtained for Arrays of Strings, or with single Strings (which will first be tokenized).

bigrams

ngrams = Cadmium.ngrams.new
ngrams.bigrams("these are some words")
# => [["these", "are"], ["are", "some"], ["some", "words"]]

trigrams

ngrams = Cadmium.ngrams.new
ngrams.trigrams("these are some words")
# => [["these", "are", "some"], ["are", "some", "words"]]

arbitrary n-grams

ngrams = Cadmium.ngrams.new
ngrams.ngrams("some other words here for you", 4)
# => [["some", "other", "words", "here"], ["other", "words", "here", "for"], ["words", "here", "for", "you"]]

padding

n-grams can also be returned with left or right padding by passing a start and/or end symbol to the bigrams, trigrams or ngrams.

ngrams = Cadmium.ngrams.new
ngrams.ngrams("these are some words", 4, "[start]", "[end]")
# => [
      ["[start]", "[start]", "[start]", "these"],
      ["[start]", "[start]", "these", "are"],
      ["[start]", "these", "are", "some"],
      ["these", "are", "some", "words"],
      ["are", "some", "words", "[end]"],
      ["some", "words", "[end]", "[end]"],
      ["words", "[end]", "[end]", "[end]"]
    ]

Classifiers

Cadmium comes with one classifier so far, a Classic Bayes classifier. It is a probabalistic classifier that, when trained with a data set, can classify words according to categories.

classifier = Cadmium.bayes_classifier.new

classifier.train("crystal is an awesome programming language", "programming")
classifier.train("ruby is nice, but not as fast as crystal", "programming")

classifier.train("my wife and I went to the beach", "off-topic")
classifier.train("my dog likes to go outside and play", "off-topic")

classifier.categorize("Crystal is my favorite!")
# => "programming"

tf-idf

Term Frequency–Inverse Document Frequency (tf-idf) is implemented to determine how important a word (or words) is to a document relative to a corpus. The following example will add four documents to a corpus and determine the weight of the word "crystal" and then the weight of the word "ruby" in each document.

tfidf = Cadmium.tf_idf.new
tfidf.add_document("this document is about crystal.")
tfidf.add_document("this document is about ruby.")
tfidf.add_document("this document is about ruby and crystal.")
tfidf.add_document("this document is about crystal. it has crystal examples")

puts "crystal --------------------------------"
tfidf.tfidfs("crystal") do |i, measure, key|
  puts "document ##{i} is #{measure}"
end

puts "ruby --------------------------------"
tfidf.tfidfs("ruby") do |i, measure, key|
  puts "document ##{i} is #{measure}"
end

# =>  crystal --------------------------------
      document #0 is 1
      document #1 is 0
      document #2 is 1
      document #3 is 2
      ruby --------------------------------
      document #0 is 0
      document #1 is 1.2876820724517808
      document #2 is 1.2876820724517808
      document #3 is 0

Transliterator

The Transliterator module provides the ability to transliterate UTF-8 strings into pure ASCII so that they can be safely displayed in URL slugs or file names.

transliterator = Cadmium.transliterator

transliterator.transliterate("Привет")
# => "Privet"

transliterator.transliterate("你好朋友")
# => "Ni Hao Peng You"

# With the string extension

"މިއަދަކީ ހދ ރީތި ދވހކވ".transliterate
# => "mi'adhakee hdh reethi dhvhkv"

"こんにちは、友人".transliterate
# => konnichiwa, You Ren

Sentiment Analysis

The Sentiment module uses the AFINN-165 wordlist and Emoji Sentiment Ranking to provide sentiment analysis on arbitrary blocks of text.

sentiment = Cadmium.sentiment

"Crystal is seriously the best programming language.".sentiment
# or
sentiment.analyze("Crystal is seriously the best programming language.")
# =>  {
        score: 3,
        comparative: 0,
        tokens: ["Crystal", "is", "seriously", "the", "best", "programming", "language"],
        words: ["best"],
        positive: ["best"],
        negative: []
      }

"I really hate Python".is_negative?
# => true

"I really 💗 Crystal. It's my favorite.".is_positive?
# => true

Tries

A trie is a data structure for efficiently storing and retrieving strings with identical prefixes, like "meet" and "meek".

trie = Cadmium.trie.new

trie.add("meet")
trie.size
# => 5

trie.add("meek")
trie.size
# => 6

trie.contains?("meet")
# => true

trie.find_prefix("meeting")
# => {"meet", "ing"}
trie.find_prefix("meet")
# => {"meet", ""}
trie.find_prefix("me")
# => {nil, "me"}

trie.keys_with_prefix("me")
# => ["meet", "meek"]

trie.add(["m", "me"])
trie.matches_on_path("meeting")
# => ["m", "me", "meet"]

Edge Weighted Digraph

EdgeWeightedDigraph represents a digraph, you can add an edge, get the number vertexes, edges, get all edges and use toString to print the Digraph.

digraph = Cadmium.edge_weighted_digraph.new

digraph.add(5, 4, 0.35)
digraph.add(5, 1, 0.32)
digraph.add(1, 3, 0.29)
digraph.add(6, 2, 0.40)
digraph.add(3, 6, 0.52)
digraph.add(6, 4, 0.93)

puts digraph.v # => 7
puts digraph.e # => 6

Readability

Analyze blocks of text and determine, using various algorithms, the readability of the text.

text = <<-EOF
    After marriage, the next big event in the couples lives will be their honeymoon. It is a time when the newly weds can get away from relatives and friends to spend some significant time getting to know one another. This time alone together that the couple shares is called the honeymoon. A great gift idea for the married couple would be to give them a surprise tour package. Most women would like to go on a honeymoon.
    The week or two before the ceremonies would be the best time to schedule a tour because then the budget for this event could be considered. In winter there are more opportunities for the couple to get close to one another because of the cold weather. It is easier to snuggle when the weather is not favorable to outdoor activities. This would afford the couple ample time to know more about themselves during the honeymoon.
    Honeymoon plans should be discussed with the wife to ensure that the shock is pleasant and not a negative experience to her. It is also a good idea in this case, to ask her probing questions as to where she would like to go. Perhaps you could get a friend or family member to ask her what would be her favorite travel location. That would ensure that you know just what she is looking for.
    Make sure that the trip is exactly what she wants. Then on the wedding night tell her about the adventure so that the needed accommodations can be made.
EOF

report = Cadmium.readability.new(text)

puts report.flesch  # => 71.47176470588238
puts report.fog     # => 10.721568627450981
puts report.kincaid # => 7.513725490196077

WordNet

WordNet® is a large lexical database of English. Nouns, verbs, adjectives and adverbs are grouped into sets of cognitive synonyms (synsets), each expressing a distinct concept. Synsets are interlinked by means of conceptual-semantic and lexical relations. - https://wordnet.princeton.edu/

This WordNet implimentation is based almost completely on doches ruby library rwordnet with some extras thrown in and, of course, backed by the speed and type safety of Crystal. This is experimental and the API may change, but WordNet brings the power of the English (and hopefully other languages in the future) dictionary to your programs.

Using it is easy with Cadmium's API.

# Lookup a single word with a specific part of speech
lemma = Cadmium.wordnet.lookup("horse", :n)
puts lemma.word.capitalize + " - " + lemma.pos
lemma.synsets.each_with_index do |synset, i|
  puts "#{i + 1}. #{synset.gloss}"
end

# Lookup a single word accross all parts of speech
lemmas = Cadmium.wordnet.lookup("horse")
lemmas = lemmas.map { |l| {word: l.word, pos: l.pos, synsets: l.synsets} }
lemmas.each do |l|
  word = l[:word].capitalize
  pos = l[:pos]
  l[:synsets].each do |s|
    puts "#{word} (#{pos}) - #{s.gloss}"
  end
end


# Lookup a definition by offset and part of speech
synset = Cadmium.wordnet.get(4424418, :n)
puts "---------------------------------------------"
puts synset.synset_offset
puts synset.pos
puts synset.gloss
puts synset.word_counts

Roadmap

This is all I want to have done before a v1.0 release.

• [x] Tokenizers
  • [x] AggressiveTokenizer
    • [x] i18n
  • [x] CaseTokenizer
  • [x] Pragmatic ?
  • [x] RegexTokenizer
  • [x] SentenceTokenizer
  • [x] TreebankWordTokenizer
  • [x] WhitespaceTokenizer
  • [x] WordPunctuationTokenizer
• [x] String Distance
  • [x] Levenshein
    • [ ] Approximate String Matching
  • [x] JaroWinkler
• [ ] Stemmers
  • [x] PorterStemmer
    • [ ] i18n
  • [ ] LancasterStemmer
    • [ ] i18n
• [ ] Classifiers
  • [x] Bayes
  • [ ] Logic Regression
• [x] Phonetics
  • [x] SoundEx
  • [x] Metaphone
  • [ ] Double Metaphone
• [x] Inflectors
  • [x] Count
  • [x] Noun
  • [x] Verb
  • [ ] i18n
• [x] N-Grams
• [x] TF-IDF
• [x] Transliterator
• [x] Sentiment Analysis
• [x] Tries
• [x] EdgeWeightedDigraph
• [x] Readability
• [x] WordNet
• [ ] ShortestPathTree
• [ ] LongestPathTree
• [ ] Spellcheck
• [ ] POS Tagger
• [ ] Word2Vec

Contributing

1. Fork it ( https://github.com/watzon/cadmium/fork )
2. Create your feature branch (git checkout -b my-new-feature)
3. Commit your changes (git commit -am 'Add some feature')
4. Push to the branch (git push origin my-new-feature)
5. Create a new Pull Request

Contributors

• watzon Chris Watson - creator, maintainer